A newly published CVE describes a code-injection flaw in ash_ai, an AI toolkit built on the Elixir Ash framework, that lets an unauthenticated remote attacker run arbitrary Elixir code. The bug stems from how the library's prompt-handling logic feeds action arguments into an EEx template evaluator, so an application that builds a prompt from user-supplied request data can end up compiling and executing attacker-controlled input. Elixir teams using ash_ai's prompt actions should treat this as a priority patch.
// source: nvd.nist.gov ↗